use currentDefaultPath instead of <path>'s d attribute, fixes stroke's different behavior in SVG and canvas.

[zenozeng]

Now __addPathCommand will only update this.__currentDefaultPath. And __applyCurrentDefaultPath will be called inside stroke or fill.

The bug:

The left image is SVG, the middle image is Canvas, and the right is diff bitmap.

The following 2 code blocks should have different result:

    ctx.beginPath();
    ctx.moveTo(10,10);
    ctx.lineTo(30, 30);
    ctx.lineTo(60, 10);
    ctx.stroke(); // currentDefaultPath not including the final Z
    ctx.closePath();

    ctx.beginPath();
    ctx.moveTo(10,10);
    ctx.lineTo(30, 30);
    ctx.lineTo(60, 10);
    ctx.closePath();
    ctx.stroke(); // currentDefaultPath including the final Z

See also: zenozeng/p5.js-svg#38

[gwwar]

Good catch! 👍 Maybe add a unit test with this case to https://github.com/gliffy/canvas2svg/blob/master/jasmine-tests/spec/canvas2svgspec.js

[zenozeng]

BTW, what about adding a bitmap diff based unit tests?
I think it's hard to tests the behavior without using pixels.

Just like this test: http://zenozeng.github.io/p5.js-svg/test/

[gwwar]

Image diffs are great for smoke tests. Nice trick with the data:image/svg+xml. It looks like canvas is happy accepting drawImage with an uri of data:image/svg+xml. http://jsfiddle.net/gwwar/n4sgo9ao/

We have had some smoke tests in the past with a clunkier set up of taking screenshots in a headless browser, and then passing that through imagemagick to get a diff, but this looks a lot simpler.

If I have time I might whip something up. Perhaps something more interactive like the example/issue reporter in canvg? http://gabelerner.github.io/canvg/examples/index.htm